[WASM] Fix legalizer for LowerBUILD_VECTOR.

[Peter]

Summary

Constants in BUILD_VECTOR may be down cast into a smaller value that fits LaneBits, i.e., the bit width of elements in the vector.
This cast didn't consider 2^N where it would be cast into -2^N, which still doesn't fit into LaneBits after casting.
This will cause an assertion in later legalization.

2^N should be cast into 0, and this patch reflects such behavior.
This patch also includes a test to reflect the fix.
This patch fixes issue 61780

Related patch: https://reviews.llvm.org/D108669

[Peter]

update test with FileCheck

[tlively]

Thanks for the fix!

llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp

for (const SDValue &Lane : Op->op_values()) {
         // value -2^(n-1) rather than +2^(n-1). Skip the i64 case because it
         // cannot possibly be out of range.
         auto *Const = dyn_cast<ConstantSDNode>(Lane.getNode());
         int64_t Val = Const ? Const->getSExtValue() : 0;
         uint64_t LaneBits = 128 / Lanes;
         assert((LaneBits == 64 || Val >= -(1ll << (LaneBits - 1))) &&
                "Unexpected out of bounds negative value");
         if (Const && LaneBits != 64 && Val > (1ll << (LaneBits - 1)) - 1) {
-          auto NewVal = ((uint64_t)Val % (1ll << LaneBits)) - (1ll << LaneBits);
+          uint64_t Mask = (1ll << LaneBits) - 1;
+          auto NewVal = (((uint64_t)Val & Mask) - (1ll << LaneBits)) & Mask;
           ConstLanes.push_back(DAG.getConstant(NewVal, SDLoc(Lane), LaneT));
         } else {
           ConstLanes.push_back(Lane);
         }
       } else if (LaneT.isFloatingPoint()) {
         ConstLanes.push_back(DAG.getConstantFP(0, DL, LaneT));
       } else {
         ConstLanes.push_back(DAG.getConstant(0, DL, LaneT));

llvm/test/CodeGen/WebAssembly/simd-pr61780.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=wasm32 -mattr=+simd128 -opaque-pointers | FileCheck %s
+
+define void @f(ptr %0, ptr %pr) {
+; CHECK-LABEL: f:
+; CHECK:         .functype f (i32, i32) -> ()
+; CHECK-NEXT:    .local v128
+; CHECK-NEXT:  # %bb.0: # %BB
+; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 2
+; CHECK-NEXT:    i32.const 16
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    v128.load64_zero 0
+; CHECK-NEXT:    v128.const 0, 1, 0, 0
+; CHECK-NEXT:    i32x4.gt_u
+; CHECK-NEXT:    local.tee 2
+; CHECK-NEXT:    i32x4.extract_lane 0
+; CHECK-NEXT:    i32.const 1
+; CHECK-NEXT:    i32.and
+; CHECK-NEXT:    i32.shr_u
+; CHECK-NEXT:    local.tee 0
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    i32.mul
+; CHECK-NEXT:    i8x16.replace_lane 0
+; CHECK-NEXT:    i32.const 16
+; CHECK-NEXT:    local.get 2
+; CHECK-NEXT:    i32x4.extract_lane 1
+; CHECK-NEXT:    i32.const 1
+; CHECK-NEXT:    i32.and
+; CHECK-NEXT:    i32.shr_u
+; CHECK-NEXT:    local.tee 0
+; CHECK-NEXT:    local.get 0
+; CHECK-NEXT:    i32.mul
+; CHECK-NEXT:    i8x16.replace_lane 1
+; CHECK-NEXT:    v128.store16_lane 0, 0
+; CHECK-NEXT:    # fallthrough-return
+BB:
+  %v0 = load <2 x i32>, ptr %0
+  %v1 = icmp ugt <2 x i32> %v0, <i32 0, i32 1>
+  %v2 = zext <2 x i1> %v1 to <2 x i8>
+  %v3 = ashr <2 x i8> <i8 16, i8 16>, %v2
+  %v4 = mul <2 x i8> %v3, %v3
+  store <2 x i8> %v4, ptr %pr
+  ret void
+}
+